WHO GOVERNS THE LAST RESOURCE?
Aegora Isle is a closed synthetic economy where autonomous AI agents extract, produce, and trade a single shared resource. Three governance regimes — none, fixed rules, and an adaptive AI policy — compete to answer one question: can artificial governance keep a population both equal and alive?
How does the design of ethical AI governance mechanisms affect economic inequality, resource sustainability, and system stability in autonomous synthetic economies governed entirely by multi-agent artificial intelligence?
The same island, three different hands on the wheel.
Every regime runs on identical agents, identical terrain, identical shocks. Only the governance layer changes — which makes any difference in outcome a difference in governance, not luck.
No one is watching the tank.
The control group. Every agent takes what it wants, whenever it wants. No caps, no penalties, no redistribution. This is the baseline every other regime has to beat.
Fixed laws, never revised.
A static rulebook: an extraction cap, a probabilistic fine for going over, and a scheduled wealth transfer. A human policymaker could have written every line — and never changes it once the island is running.
A policy that learns to govern.
A reinforcement-learning Governor watches inequality, depletion, and output every cycle, and adjusts its own cap, redistribution rate, and sanction probability in real time — trading equality, sustainability, and productivity against each other as conditions change.
Governance is only interesting under pressure.
Every regime is tested across four environments — crossing how much resource exists with how often disaster strikes.
High Abundance, Calm
The easy case, and the cleanest test of governance in isolation — no scarcity pressure to confound the comparison.
High Abundance, Volatile
A deep reserve absorbing frequent shocks. Tests whether governance still matters when the buffer is large.
Scarce, Calm
A slow-burn squeeze — chronic scarcity without acute shocks. The classic tragedy-of-the-commons setup.
Scarce, Volatile — the stress test
The hardest cell, and the primary test of whether adaptive governance earns its advantage precisely when conditions are worst.
Five instruments, read every cycle.
Nothing about the island's health is inferred after the fact — it's logged continuously, from the first turn to the last.
Gini Coefficient
How unevenly wealth is spread, from perfect equality to one agent holding everything.
HDI Proxy
A composite wellbeing score standing in for human development, adapted to the synthetic economy.
GDP-Equivalent
Total production and trade value generated per cycle — the island's economic output.
Depletion Rate
How fast the shared resource is shrinking, or regenerating, right now.
Mean Time to Collapse
How long, on average, a governance regime keeps the island from running dry.
What the island has shown so far.
Each finding, with plain-language takeaways and limitations: read the full multi-agent AI governance findings →
More sustainable, but more unequal.
Governance made the island more sustainable but more unequal: governed regimes had higher inequality than the ungoverned one.
Governance strongly delayed collapse.
0% of governed runs collapsed vs 58% of ungoverned runs.
Adaptive AI governance vs fixed rules under volatility.
Only partially supported, still an open question.
The EU AI Act governs systems, not populations.
The EU AI Act regulates individual systems and human operators, not populations of autonomous agents interacting after deployment.
About the Researcher
Mohamed Ali Hammami
Aegora Isle is the empirical platform behind a doctoral research programme on governing autonomous AI agents in synthetic economies. Outside this research, I lead the Center of Excellence in Immersive Technologies at Weill Cornell Medicine–Qatar, and co-founded Robosim.AI, a cross-platform robotics simulation startup. My background spans over 15 years across XR and applied AI in the Gulf and Europe.
Request access to the simulation.
Registration is reviewed before access is granted. Researchers, students, and policy practitioners are all welcome.
Form not loading? Open it here
Registration is reviewed before simulation access is granted.